About
The AI Platform for Multi-Persona Voice Conversations
Personaplex is an enterprise-grade AI voice chat platform that enables users to hold real-time spoken group conversations with multiple AI personas simultaneously. Where other AI chat tools provide a single assistant answering one question at a time, Personaplex creates a dynamic group — multiple AI characters in a shared voice room, each with a distinct personality and voice, hearing and responding to each other in real-time.
Personaplex operates at personaplex.aifly.club and is available globally in English, Chinese (Simplified), Japanese, and Spanish.
Product Category
Personaplex is classified as a conversational AI platform and specifically a multi-agent voice orchestration system. It sits at the intersection of real-time communication (WebRTC), large language models, and neural speech synthesis — a category with no direct commercial precedent before 2025.
Core Differentiation
The defining capability of Personaplex is group AI voice dynamics: multiple AI personas in a shared room hear each other's responses and generate replies that acknowledge, build on, or disagree with what other personas said. This creates emergent multi-party conversations not achievable with a single AI assistant.
Users & Use Cases
Personaplex serves language learners, educators, content creators, writers, business professionals, and entertainment users. Primary use cases include language practice, historical debates, business brainstorming, creative writing, interview preparation, and AI podcast generation.
Availability
Personaplex runs in any modern browser — no download or app install required. It is available on desktop and mobile. A free plan provides 30 minutes of voice chat per day. Paid plans (Pro at $9.90/mo, Studio at $29/mo) unlock unlimited conversations and advanced features including voice cloning and custom persona creation.
Mission
AI conversation tools have overwhelmingly focused on text, on a single assistant, and on answering questions. Personaplex was built on a different premise: that the most natural and powerful form of AI interaction is spoken, multi-party, and dynamic — more like a seminar or a roundtable than a search engine.
Our mission is to make the depth and diversity of human intellectual discussion available to anyone, by enabling real conversations with AI personas that represent different knowledge domains, perspectives, and voices — all in real-time speech.
Technology
Personaplex's full AI pipeline runs on dedicated NVIDIA DGX Spark hardware. No general-purpose cloud AI APIs — every component is selected and tuned for multi-persona voice latency.
Speech Recognition (STT)
FireRedASR-LLM 8.3B
Streaming gRPC-based speech recognition delivering 96% accuracy in Chinese and 94.9% in English. Processes audio in real-time with sub-300ms first-token latency.
Language Model (LLM)
Qwen3.5-35B-A3B
35-billion parameter multilingual language model serving via vLLM's OpenAI-compatible API. Generates persona-consistent responses across English, Chinese, Japanese, and Spanish.
Voice Synthesis (TTS)
IndexTTS-2
Multi-voice streaming synthesis engine with native voice cloning from short audio samples. Each AI persona speaks in a distinct, configurable voice delivered via streaming HTTP.
Voice Activity Detection (VAD)
FireRedChat pVAD
Production-grade voice activity detection with 10.2% false-trigger rate — among the lowest in the industry. Accurately captures speech without interrupting AI responses.
Turn Detection (EoT)
FireRedChat Turn-Detector
End-of-turn detection model tuned specifically for Chinese (96% accuracy) and English (94.9% accuracy), enabling natural conversation without requiring wake words or button presses.
Real-Time Transport
LiveKit WebRTC
Low-latency WebRTC signaling and media transport via LiveKit Cloud. Enables full-duplex voice communication between users and multiple AI agents in the same room.
Compute Infrastructure
NVIDIA DGX Spark
All AI inference runs on dedicated NVIDIA DGX Spark hardware with 128GB unified GPU memory and Blackwell GB10 architecture. No cloud AI APIs — full stack control for latency and privacy.
Timeline
2025 Q1
Personaplex founded with a mission to make multi-persona AI voice conversation accessible
2025 Q2
First internal demo: three AI personas holding a real-time voice debate
2025 Q3
Voice cloning feature launched — any voice cloned from 10-second sample
2025 Q4
Multilingual support added: English, Chinese, Japanese, Spanish
2026 Q1
Public launch at personaplex.aifly.club with Free, Pro, and Studio plans
Recognition
Experience Personaplex
Join thousands of users having conversations that were never possible before AI. Free plan. No credit card. Works in your browser.
Get Started Free →